Table 4.

Proportion of students whose behavior at follow-up we predicted correctly with different models.

	Initiation				Recent sex				Sex without a condom
	OLS (joint)	Lasso (joint)	Lasso (joint with implicit main effects)	OLS (separate)	OLS (joint)	Lasso (joint)	Lasso (joint with implicit main effects)	OLS (separate)	OLS (joint)	Lasso (joint)	Lasso (joint with implicit main effects)	OLS (separate)
Total	0.806	0.801	0.808	0.811	0.812	0.813	0.854	0.846	0.848
Boys	0.773	0.752	0.780	0.798	0.784	0.795	0.799	0.825	0.843	0.812	0.818	0.849
Girls	0.820	0.828	0.822	0.835	0.829	0.822	0.821	0.834	0.862	0.870	0.870	0.877
Black	0.803	0.795	0.802	0.808	0.806	0.805	0.807	0.818	0.858	0.847	0.848	0.856
All other races	0.814	0.820	0.831	0.913	0.827	0.837	0.834	0.876	0.840	0.840	0.847	0.911
Grade 9	0.810	0.807	0.822	0.827	0.836	0.834	0.832	0.857	0.900	0.887	0.891	0.907
Grade 10	0.801	0.793	0.791	0.801	0.792	0.796	0.798	0.818	0.820	0.817	0.821	0.856
LGB	0.833	0.833	0.833	.	0.794	0.794	0.778	.	0.841	0.825	0.810	.
Non-LGB	0.804	0.799	0.807	.	0.812	0.813	0.815	.	0.855	0.847	0.850	.

Note: OLS (joint) is the OLS model estimated on all students. Lasso (joint) is the traditional lasso model estimated on all students. Lasso (joint, with implicit main effect) is the lasso model with the addition of all baseline variables that were selected in the interaction terms. OLS (separate) are OLS regressions stratified by the demographic characteristic listed on the left. OLS = Ordinary least squares; LGB = Lesbian, gay, and bisexual.